Agile Software Development:
what can we learn as researchers?
Pierre Lison,
Language Technology Group (LTG)
Department of Informatics
LTG Tech Lunch
October 31 2012
Introduction
As NLP researchers, we spend a lot of our
time dealing with code
(Reading/designing/writing/debugging/testing, etc.)
2
Introduction
• In many ways, the quality of our code has a
decisive impact on the quality of our research
1. Good code → (often) better empirical results
2. Good code often helps us get a better understanding
of our research problem (concepts, limitations, etc.)
3. Good code is easier to extend, reuse and refactor in
several experiments or projects
4. Good code makes it easier for other researchers to
understand our work, and adopt it in their own research
3
Introduction
• ... but strangely enough, we (researchers)
rarely reflect on the adequacy of our
development methods
•
•
•
•
Are our development methods optimal?
•
Do our methods promote or hinder collaboration?
Do we focus on the right (=high priority) aspects?
Do we control the quality of our code?
How do we deal with unexpected events (e.g.
unforeseen problems, change in approach)?
4
Introduction
• Software engineering changed
a lot in the last 10 years
• Agile development methods
increasingly popular
• I would like to talk about
some of these new ideas
•
And most importantly, how they can
help us do better research
5
Waterfall model
Traditional
way of
building
software
Requirements
Design
Implementation
«Big Design
Up Front»
Testing
Maintenance
6
Problems with the waterfall
• Drawbacks of the waterfall model:
•
•
The software requirements often vague & volatile
•
•
•
Working software only available at the latest stages
Many design issues only become apparent at
implementation time
Inability to adapt to unforeseen events
Typically leads to rigid division of labour
Alternative: develop software in a
more incremental & iterative fashion
7
Incremental development
Requirements
Requirements
Requirements
Design
Design
Design
Implementation
Implementation
Implementation
Testing
Testing
Testing
Testing
Testing
Implementation
Implementation
Design
Design
Requirements
Requirements
Iteration Iteration Iteration Iteration
1
2
3
4
...
8
Incremental development
• Idea: construct the software via a
sequence of several short iterations
•
Iteration purpose is to integrate a new functionality
(the one with the highest priority at the moment)
•
Each iteration includes some basic requirements
analysis, design, implementation and testing
•
At the end of each iteration, we have a working
system, extended with the given function
9
Incremental development
One particular type of Agile method: Scrum
10
Incremental development
• Advantages of incremental & iterative
development methods:
•
Fast delivery of a working system, even though it may
be imperfect or incomplete
•
Gradual refinement and extension of the software
requirements and system design
•
Greater adaptivity to unforeseen changes
(implementation problems, external events)
11
Incremental development
Waterfall
predictive planning
Agile
adaptive planning
Better when uncertainty is high:
• unclear or changing «requirements»
• technological risks
• social/organisational factors
• may only achieve a subset of goals
High-uncertainty is the norm in academic research
12
Agile: a lightweight methodology
Agile put the emphasis on
working software as the core
development objective
Project plans,
Requirement specs,
Design diagrams,
Progress reports
Gantt charts
etc.
have no values
in themselves
should only be
done if they help
the development
process
13
Agile: a lightweight methodology
• Organisational structure is also lightweight
•
•
No rigid roles or hierarchy, work is largely self-directed
•
Emphasis on direct, face to face collaboration
Users seen as partners directly engaged in the
development process
14
Agile manifesto
[The Agile Manifesto: http://agilemanifesto.org/]
15
Agile methodologies
Scrum
Crystal
Agile Unified Process
Lean
Extreme Programming (XP)
16
Good practices
Four agile engineering practices:
Unit testing
Refactoring
Write systematic test cases for every
unit of code to ensure the
requirements are satisfied. Run the
unit tests after any code change.
Modify the code’s internal structure
(without altering its behaviour) to
follow standard patterns, increase
readability and extensibility
Test-driven
development
Continuous
Integration
Write test cases first, as a way to
define the software requirements.
Then use these tests to control the
development progress
Commit written code to repository
and rebuild system as soon as
possible. Automatically control for
integration problems.
17
Conclusion
• Agile development
methodologies can help
us write better code
•
Improved quality, faster
delivery, increased flexibility
•
Especially useful for research
systems, which must typically
face high uncertainty
•
Lightweight, but highly
disciplined methodology!
18
Conclusion
• Key ideas
•
Development as a sequence of
short iterations gradually
extending or improving the
system
•
Working code is the primary
focus, not procedures or
hierarchical roles
•
Adaptivity: Embrace change
instead of trying to predict it
19